Motivated by studies demonstrating the influence of the synoptic- and meso-α scales upon deep, moist convection, the Mesoscale Predictability Experiment (MPEX) hypothesizes that the collection of non-routine synoptic- and meso-α-scale observations in the upstream, pre-convective environment across the Intermountain West and their subsequent assimilation into convection-permitting numerical forecasts significantly improves forecasts of the timing, location, and mode of convection initiation and evolution downstream. Herein, utilizing output from two thirty-member ensembles of Ensemble Kalman filter-initialized, convection-permitting real-data numerical simulations, one incorporating MPEX observations and one not, we test the related hypothesis that a more intensively-sampled representation of the pre-convective atmospheric state is sufficient to improve the practical predictability of initial (or pristine) CI timing and location over the range of fifteen events sampled by MPEX between 15 May and 15 June 2013.
Observed and simulated radar reflectivity data exceeding 35 dBz at the -10 C level is used to identify observed and numerically-simulated initial CI events for each MPEX event. Subsequently, probabilistic, temporally-binned spatial verification methods are utilized to examine forecast skill. For a given event, the timing and location of the initial observed and simulated (by each ensemble member) occurrence of CI are first identified. The probability of CI occurring within S km of each model grid point within a T h window is defined as the number of ensemble members within each ensemble correctly forecasting CI within these thresholds divided by the total number of ensemble members. Spatiotemporal thresholds considered are 40 km/1 h, 80 km/1.5 h, 120 km/2 h, 160 km/2.5 h, and 200 km/3 h. The forecast skill of each ensemble is evaluated at these spatiotemporal thresholds using a Brier Skill Score (BSS) computed with reference to the sample climatology. Results from one or more MPEX events will be presented. Connections to previous deterministic and probabilistic investigations into CI predictability will be drawn as appropriate.